Improving compound–protein interaction prediction by building up highly credible negative samples

• DOI: 10.1093/bioinformatics/btv256
• Publication Date: 2015-06-13T17:12:36Z
• Authors: Hui Liu; Jianjiang Sun; Jihong Guan; Jie Zheng; Shuigeng Zhou
• Keywords: 0301 basic medicine; Support Vector Machine; Ismb/Eccb 2015 Proceedings Papers Committee July 10 to July 14, 2015, Dublin, Ireland; Proteins; Bayes Theorem; 03 medical and health sciences; Pharmaceutical Preparations; Drug Discovery; Protein Interaction Mapping; Humans; Computer Simulation; Amino Acid Sequence; Databases, Protein

Abstract Motivation: Computational prediction of compound–protein interactions (CPIs) is great importance for drug design and development, as genome-scale experimental validation CPIs not only time-consuming but also prohibitively expensive. With the availability an increasing number validated interactions, performance computational approaches severely impended by lack reliable negative CPI samples. A systematic method screening sample becomes critical to improving in silico methods. Results: This article aims at building up a set highly credible samples via method. As most existing models assume that similar compounds are likely interact with target proteins achieve remarkable performance, it rational identify potential based on converse proposition dissimilar every known/predicted compound much be targeted vice versa. We integrated various resources, including chemical structures, expression profiles side effects compounds, amino acid sequences, protein–protein interaction network functional annotations proteins, into framework. first tested screened six classical classifiers, all these classifiers achieved remarkably higher our than randomly generated both human Caenorhabditis elegans. then verified three models, bipartite local model, Gaussian kernel profile Bayesian matrix factorization, found performances significantly improved Moreover, we bioactivity dataset. Finally, derived two sets new training support vector machine classifier positive annotated DrugBank interactions. The predicted provide research community useful resource identifying targets helpful supplement current curated databases. Availability: Supplementary files available at: http://admis.fudan.edu.cn/negative-cpi/. Contact: [e-mail] Information: data Bioinformatics online.